TCEC superfinal, what will it be like?

[S.Taylor]

What upgrades will there be different to stage 3?
And how are they comparable with the same programs in TCEC season 7?

I'm sure SF has made corrections and maybe upgrade of its playing strength. Even if no upgrade in elo, I feel the original SF at the beginning of stage 3, but with clock corrected, may be stronger than Komodo 9.2.

Or is Komodo also submitting a further improved version again?

[JJJ]

Stockfish regression test showed +60 ELO from Stockfish 6. I think they improved at least 15 elo recently from dev version and probably a little more with Lazy SMP patch on high number of core.

I don't know how better is now Komodo, so It is tough to say who is favorite. I need to know more about Komodo dev to be sure. And of course, it is a better Komodo who will play.

[Laskos]

Stockfish regression test showed +60 ELO from Stockfish 6. I think they improved at least 15 elo recently from dev version and probably a little more with Lazy SMP patch on high number of core.

I don't know how better is now Komodo, so It is tough to say who is favorite. I need to know more about Komodo dev to be sure. And of course, it is a better Komodo who will play.

These are a bit another ELOs, ultra-bullet and self-games, more like 30-45 ELO points at 5+3 and varied opposition. Roughly similar amount to Komodo improvement during the same period of time. Lazy_SMP at TCEC is a wild card, though. Let's see how it performs.

[S.Taylor]

Now at this very momment, we are awaiting the final stage, "superfinal".
No exact details have been posted.
(I suppose they are much too busy rushing to be ready!)

[Eelco de Groot]

Now at this very momment, we are awaiting the final stage, "superfinal".
No exact details have been posted.
(I suppose they are much too busy rushing to be ready!)

From Fishtest:

Martin stated that the superfinal will start on Nov. 3rd regardless of the finish date of stage 3, so we have still a week.

Yes, at the moment Stockfish is being completely rewritten and maybe even set to autocompile. Komodo team will be entering a version they have been secretely testing in their beta team. So they have the advantage here.

I am just kidding, I hope you understand. None of the above is true I do agree with Kai, from a theoretical perspective, Lazy SMP is really just LAZY SMP and should not perform better. Disclaimer: that would be my first reaction without looking at any specific code, so just for the argument here. But then the question is, is Komodo using still Don's version of Lazy SMP (basically I mean, because it was never really finished by Don) and is that why it is so strong at many threads? I don't know. If Lazy SMP for Stockfish means it is searching 'wider', especially at the root then this may be explaining the advantage in that it is correcting the fact that at the very high depths achieved in TCEC you don't really need the Singular Extensions in the PV anymore, or that they are set to extend too much at these depths. Combined with an oversupply of threads for an serial algorithm like alpha beta. Lazy SMP would correct some of that, but in a 'lazy' way. I would expect, Stockfish' Lazy SMP not to be much beter than Toga's version of Lazy SMP: reasonable, but not scaling spectacularly well especially above four threads...

[Joerg Oster]

Now at this very momment, we are awaiting the final stage, "superfinal".
No exact details have been posted.
(I suppose they are much too busy rushing to be ready!)

From Fishtest:

Martin stated that the superfinal will start on Nov. 3rd regardless of the finish date of stage 3, so we have still a week.

Yes, at the moment Stockfish is being completely rewritten and maybe even set to autocompile. Komodo team will be entering a version they have been secretely testing in their beta team. So they have the advantage here.

I am just kidding, I hope you understand. None of the above is true I do agree with Kai, from a theoretical perspective, Lazy SMP is really just LAZY SMP and should not perform better. Disclaimer: that would be my first reaction without looking at any specific code, so just for the argument here. But then the question is, is Komodo using still Don's version of Lazy SMP (basically I mean, because it was never really finished by Don) and is that why it is so strong at many threads? I don't know. If Lazy SMP for Stockfish means it is searching 'wider', especially at the root then this may be explaining the advantage in that it is correcting the fact that at the very high depths achieved in TCEC you don't really need the Singular Extensions in the PV anymore, or that they are set to extend too much at these depths. Combined with an oversupply of threads for an serial algorithm like alpha beta. Lazy SMP would correct some of that, but in a 'lazy' way. I would expect, Stockfish' Lazy SMP not to be much beter than Toga's version of Lazy SMP: reasonable, but not scaling spectacularly well especially above four threads...

Well, test results just showed the opposite.
It was difficult to get en par with YBW performance and 3 threads. While Lazy SMP gave better performance right from the beginning with a higher number of threads, iirc.

[lucasart]

What upgrades will there be different to stage 3?
And how are they comparable with the same programs in TCEC season 7?

I'm sure SF has made corrections and maybe upgrade of its playing strength. Even if no upgrade in elo, I feel the original SF at the beginning of stage 3, but with clock corrected, may be stronger than Komodo 9.2.

Or is Komodo also submitting a further improved version again?

>80% draws. You can trust Cato for that

[whereagles]

Cato promised a bloodbath for the SF (superfinal lol). As in, unbalanced dynamical positions in the ~0.40-0.50 eval range.

But keep in mind it's not him choosing the openings this time.

[mjlef]

I have seen amazingly good results of Komodo 9.2 versus the latest Stockfish. And I have seen equally bad results of newer Komodo versions versus the same Stockfish. And we have good results of the new Komodo versus Komodo 9.2. But no results at all on a tcec-like machine at superfinal time controls. Overall, I make no predictions about the super final.

OK, back to vacation!

Mark

[whereagles]

It's been a long time since season 7 and a lot has happened. K and SF both had many changes.

Hard to call a favourite, but I'd still put my money on Komodo... (which by the way should be renamed to a much more stylish "Dragon" ).